Stochastische Überlagerung mit Hilfe der Mischungsverteilung

The paper considers the effect of additive and multiplicative measurement errors on the estimation of linear models.We assume that such measurement errors have been applied to the micro data by purpose in order to protect them against re-identification. In particular measurement errors with a bimodal mixture distribution are analyzed. First the case of cross-section data is assumed. Then for panel data both the "naive' estimator ("within estimator", mixed effects estimator) and IV estimators are considered. In particular the effect of autocorrelation of regressors in short panels is discussed.